In practical production and life, character locating and recognition has many applications, such as recognition of a vehicle license number, a banknote crown word number and an identity number in an identity card, and detection of printed numbers in various product packaging. As these characters have significance or a commercial value, it is more and more important to automatically recognize them in practical production and life.
Subject to factors such as a shooting device, an illumination condition and a shooting scene, a captured image usually has a complex background an unclear target region, thus effective image segmentation can not be performed.
The character locating is a critical precondition of recognition, and a locating precision thereof directly influences an accuracy of the recognition. In most conventional methods, a target locating is performed after an image is converted into a binary image. The disadvantage of this kind of methods is that the locating precision relies on an effect of threshold segmentation. A target locating error caused by a poor binarization effect will occur if the image has a complex scene, a low contrast or an uneven illumination. FIG. 1 is an image of an optical character region in a banknote, and the banknote is klunky due to a long time of circulation, which severely degrades an image quality and results in a very complex character background, thus it is difficult to extract characters from the image. And the characters can not be effectively extracted after the image is converted into a binary image, where the effect after the converting is illustrated in FIG. 1.1.
The character locating can be performed after an image is converted into a binary image, and the object of locating is achieved with projection information of the binary image in the horizontal direction and the vertical direction in a conventional method. The limitation of the method is that an algorithm thereof relies on the binarization process. Rupture or adhesion of characters usually occurs after binarization in a case of a complex target background or a poor shooting effect. Under this circumstance, the projection information is inaccurate, which influences an effect of the locating. A result illustrated in FIG. 1.2 is obtained by processing FIG. 1.1 with a method of projection. Curves C1 and C2 in FIG. 1.2 are projection curves of the image in the horizontal direction and the vertical direction, respectively, and as can be seen, the method of projection can not effectively determine a boundary of the characters.